Method unchanged since Sep 29, 2026 · what changed then · full methodology below
This feature presents conditional market history — where each measure sits against its own past, and what individually-listed price paths followed its extremes. It makes and tests no predictive claim; no validated status exists for it. We present the record; we do not assert predictive skill.
Shiller CAPE. Price against the ten-year average of real earnings (Shiller's cyclically adjusted P/E) — earnings power.
Buffett Indicator. Total US market capitalization against GDP — economic size.
Household Equity Allocation. The share of household financial assets held in equities — positioning.
Excess CAPE Yield. The CAPE earnings yield minus the ten-year real interest rate — what stocks yield over bonds after inflation, on Shiller’s own convention. Low is expensive.
Four lenses because they cover four mechanisms — earnings power, economic size, positioning, and the margin over bonds. The trailing P/E was examined and excluded: it inverts at earnings collapses — it read roughly 120 at the 2009 lows — the failure CAPE was constructed to fix. Excess CAPE Yield (ECY) arrived last and by the route the charter requires. It was chartered by amendment (VR-DR-6) and shipped in July 2026 as the fourth lens, with its own sourcing arc, its own register and its own methodology.
Each reading is ranked only against the measure’s own history up to that month — an expanding window: no later month enters a month’s rank. Ranking starts after a 120-month burn-in (40 quarters for the quarterly allocation share). The expensive tier is the 95th percentile and above; the cheap tier is the bottom fifth. The asymmetry is deliberate: under an expanding window the market sat in CAPE's top fifth for roughly 33 of the last 35 years — a top-quintile register is a register of the permanent; extremes are episodic, so the expensive level sits at p95. A period needs at least two consecutive months (quarters for the allocation share) in tier; one month out ends it — a visual cluster on the chart may be several register rows. The values themselves are today’s data, and a period moves only if the values under it do. For CAPE and earnings yield, which revise little, periods move only in their recent months. The Buffett lens divides each day by the GDP published by that day, at its value then — first-published from Dec 1991; before that, today's revised GDP stepped forward on its release lag, since no earlier first prints are archived. The allocation lens’s history is the Fed's revised accounts: first prints exist only from 2019, and three of its expensive periods (1967, 2007 and 2019) appear only on data revised since. Those periods are marked ‡ in its register. The shaded periods and the tables are defined by the percentile basis. Sigma is a reading beside them, not a second set of periods.
The CAPE trend card. The CAPE page's second today-card reads the stored CAPE trend-deviation series: how far today's CAPE sits from its own expanding log-linear trend, in residual sigmas. That is the VR-DR-4-C fit, recomputed nightly by the site's pipeline, never at page load. The reading is classified on the house five-rung ladder with boundaries at ±1σ and ±2σ. The percentile answers how expensive versus all history; the trend sigma answers how stretched versus where the trend expected it today — two questions, two answers.
The percentile ranks today against every reading in the measure's history. Sigma measures today against the measure's own fitted trend, on Grantham's basis. The trend is log-linear for CAPE and the Buffett indicator and linear for the allocation share, both fit only on the months up to and including the current one. The bases can disagree — today CAPE is higher than 99% of its own history, yet less than two sigma above its own trend — and the disagreement is information: the percentile asks 'high versus all of history?', sigma asks 'stretched versus its own path?'. Sigma readings in each measure's first twenty-four eligible periods are numerically unstable while the trend line is still settling, and are excluded from the historical summaries. The sigma shown on individual indicator pages is a different instrument: distance from the all-history average, with no trend. For a measure that drifts upward across decades, it reads far higher than the distance from the measure's own fitted path shown here. The basis is Grantham's; the settings are ours and stated — an expanding window including the current reading, log-linear for the ratio lenses and linear for the allocation share. So trend-deviation figures published elsewhere, fit with hindsight or under other settings, may differ.
CAPE never entered its bottom fifth at the 2009 trough — at roughly 13 against a 128-year history reaching single digits, it read near its own median in real time. Buyers in March 2009 did historically well anyway, and that miss is the most documented limitation of the lens. The Buffett indicator's single cheap period is 2008–09 — partly because its history only begins in 1989. No single lens is the instrument; the panel is. Level definitions are never adjusted to make a known-good date qualify.
Valuation history begins in 1881; the site’s daily S&P price tape begins in December 1927. The register’s path columns originally stopped at that price floor — fifteen pre-1928 periods were listed without outcomes. The total-return series (monthly, 1871→) removed the reason: every period in the register now carries its full outcome path, on the same total-return conventions as every other row (VR-DR-8, August 2026). The daily price floor still governs only what the price chart can draw.
This is the Valuations page's validation record, kept with its methodology. The Validation page sets it beside every other feature's.
The question. Across each lens's whole record, does where it sits on its own ladder say anything about what followed? Every eligible month is part of exactly one period — expensive, cheap, or neither. Each period is measured at one year, three years and five from the month its first reading was published. A month's CAPE is published the first day after the month, a Z.1 quarter at the Fed's release about seventy days after it, the Buffett lens the day itself, and the earnings-yield lens with the month's CPI.
"Resolved badly" was defined before anything was computed — the next twelve months negative. Five other definitions were counted first and are recorded with the test.
At one year the answer is nothing on three lenses, and a tilt on the fourth. On CAPE it is worse than nothing: expensive periods returned a median 15.2% against 9.6% for every other period, and were negative LESS often (6 of 21, against 16 of 47). On the earnings-yield lens, read from the month each period's reading was published, expensive periods came in below the rest 24 times out of 33 — a median of 12.8% against 18.2%. A page that warned about valuation on a one-year horizon would be contradicted by its own record on CAPE, the longest lens.
At three years it is not nothing. Expensive CAPE periods came in below the rest of the record's median 16 times out of 20 — a median of 21.4% against 41.2%, and on the earnings-yield lens 23 out of 30. That is the whole of the finding, and it is a tilt, not a rule: plenty of expensive periods did well.
What the finding does NOT rest on. Not on how often expensive periods LOST money over three years. Those losing windows are a handful of clustered events — the same concentration that disqualified a candidate on the Corporate Profit Share — and any rule built on them fails when a single cluster is removed. The finding rests on the counts above, which survive dropping any whole episode.
Nothing else marks the bad periods either. Fifteen conditions were tested on each lens against a bar fixed before any was computed. They covered the four-stage regime, the inflation overlay, the stress signal, Forward Stress, Economic Breadth, Housing, the profit share's own level, the lens's level and the period's length. Across the four lenses, 12 of the 39 testable candidates pointed the wrong way.
The limits, stated. Two lenses cannot answer at all: the Buffett indicator has 19 periods in its record and the household-allocation lens 24, and each page says so where its register is. Most of CAPE's record predates the engines it was tested against — the four-stage regime's record begins in 1981 and CAPE's in 1881 — so those tests reach only its recent quarter. And at five years CAPE's gap does not survive dropping the 2003–2007 cluster.
Measured September 2026 on every period in every register. This page reads other engines to DESCRIBE the record and never to classify: no reading here sets a level, opens or closes a period, or changes what is in a register. The lenses are computed from their own series and the S&P, and nothing else.
Paths are S&P total return · Shiller monthly convention (dividends reinvested monthly). The chart above remains the price tape. "Worst after" is the deepest monthly total-return level in the 36 months after each period — month-end grain, no intra-month lows. Construction and fidelity pins: "The total-return basis" on the Confluence methodology. Historical record, not a forecast.
With each data update, the registers extend mechanically: open horizons resolve, newly completed periods append, and the forward-path tally is re-taken. Nothing is regraded — there are no calls to grade.